Coworking, Private Office, or Lease? How to Match Workspace to Actual Demand

The workspace commitment question usually arrives as a headcount question. We have eleven people in Berlin now, should we get them an office? It is the wrong input, and the reason is visible once you can watch what a company's demand in a city actually does over time rather than what its org chart says.

We tracked 883 company-city relationships with at least twelve months of observable history, following each one month by month from its first booking. What the data shows is not a maturation curve. Demand in a market declares itself almost immediately, and most of it declares that it is not going to persist.

What happens to a company's demand in a new city?

Mostly it stops. Measuring the share of relationships still booking in or after a given month, each with its own eligible denominator:

Nearly half of all company-city relationships never produce a second month of activity. The curve then flattens hard: it costs 14 percentage points to get from month 1 to month 6, and only 12.5 more to get from month 6 to month 24. A market that survives six months mostly keeps going. The decision risk is concentrated in the first two quarters, which is precisely the window in which most office conversations happen.

The shape of demand is just as lopsided. Across 6,368 active market-months, the median is 3 booking days, and 32.8% of them contain exactly one booking day. In the first year, 52.9% of relationships are active in exactly one month, and 64.1% involve exactly one person. The single most common workspace market, by a wide margin, is one employee booking one day, once.

How much demand ever becomes weekly?

We define sustained weekly demand behaviourally rather than by product: three consecutive calendar months with four or more booking days each. That is the pattern a dedicated desk or a small private office would actually serve.

14.0% of 883 company-city relationships ever reach it. And the timing is the surprise.

The median onset month is 1. More than half of the markets that ever become weekly are weekly within their first two months on record. The mental model of a market that starts with occasional drop-ins, builds into a regular rhythm, and eventually earns an office describes a small minority of cases.

Does workspace usage evolve from coworking to meeting rooms to an office?

Almost never, at least not within the first year. Tracking the workspace types each relationship used in month one against the set it used across the whole year:

Only 1.1% of relationships added a second workspace type during their first year. The escalation ladder that workspace strategies are usually built around is not a description of behaviour, it is an assumption. Markets stay structurally what they were when they started.

This also corrects something worth being explicit about. Earlier Croissant research noted that markets which ever booked a meeting room were far more likely to be sustained, and it is tempting to read that as an early signal to watch for. It is not. Restricting the test to the first 90 days, only 7 of 883 markets booked a meeting room that early, and their sustained-weekly rate was 14.3% against a 14.0% base rate. That is a lift of 1.0, meaning no signal at all. A meeting-room booking is a consequence of a market having become important, not a leading indicator that it will.

What actually predicts persistent demand

Two things, both countable within 90 days: how many separate days someone booked, and how many different people booked. Each row compares the sustained-weekly rate among markets meeting the condition against those that do not, from a 14.0% base rate.

Note the second column of each pair. A market that cannot produce two booking days in 90 days reaches weekly demand 1.3% of the time. That is the most decision-useful number here, because it is a reliable negative: the absence of early activity is far more informative than its presence.

Combining the two signals gives a usable rule. The trade-off is between how confident the rule is and how many of your eventual weekly markets it catches:

The first row is the best general rule. Three booking days and two people in 90 days is a low bar, it flags about a fifth of markets, those markets reach weekly demand nine times more often than the rest, and it still captures 71.8% of every market that eventually gets there. The strictest rule is more confident but misses more than half of your real opportunities, which is the wrong error for a decision about where to give people access.

Six demand patterns, and what each one justifies

Grouping the 883 relationships by how their first year actually behaved:

Three quarters of all company-city relationships fall into the first three rows, where any fixed commitment would be a mistake. The bottom two rows, 10.5% combined, are where dedicated space becomes a genuine question. Notice that even the persistent-but-light pattern only reaches weekly demand 12.5% of the time: a market can be reliably active for a year and still not need a desk of its own.

Why even weekly demand should not buy a long lease

Of the 124 relationships that reached sustained weekly demand, 92.7% later spent three consecutive months below it. Weekly demand is a phase far more often than it is a steady state.

That figure needs an honest caveat, because it counts a market going quiet and a market ending in the same bucket: a company that stopped using Croissant in that city looks identical to one whose demand declined. So read 92.7% as an upper bound on decline and a fair bound on durability. Either way the decision implication survives. If you size a commitment to a weekly peak, the base rate says that peak is unlikely to hold for the length of a conventional lease, and the appropriate instrument is one you can exit.

A practical market evaluation checklist

1. Give access first, always. On-demand day access costs nothing when unused, and 45.2% of your markets will turn out to be a single person booking once or twice.
2. Wait 90 days and count two things. Distinct booking days and distinct people. Do not count headcount in the city, and do not count requests.
3. Under 2 booking days in 90 days: stop. These markets reach weekly demand 1.3% of the time. Keep day access on and revisit in a quarter.
4. 3 or more days and 2 or more people: watch closely. This group reaches weekly demand 45.6% of the time. Add meeting rooms here, because collaboration is what brings people back, and rooms are cheap relative to space.
5. 8 or more days and 3 or more people: model dedicated space. 74.7% of these reach weekly demand. Compare a private office against pooled access on cost per booked day, not on headcount.
6. Re-test every quarter, and prefer instruments you can exit. Given that 92.7% of weekly markets later fall below weekly, the commitment that matches this data is a flexible office you can leave, not a lease you have to assign.

The one instrument this data does not support at any level of demand is a multi-year lease signed on the strength of a headcount forecast. Local headcount does not appear anywhere in the predictors above, and the earlier finding that 52.9% of markets are active in only one month of their first year is the reason. The number of people you employ in a city tells you how many people could book. It says nothing about how many will.

Methodology and limits

The unit is a company-market relationship: one company's usage in one metro area. 949 such relationships exist in the data, and 883 have at least twelve observable months, which is the population used for every first-year and prediction figure. Survival figures use all 949 with a shrinking eligible denominator, so a relationship only counts once the relevant month is observable. A booking day is a calendar day with at least one booking. Cancelled bookings and sessions longer than 24 hours are excluded. The window ends 1 September 2026.

Sustained weekly demand means three consecutive calendar months with four or more booking days. That definition is behavioural on purpose. Private office booking is a very recent product on the platform, too new in this window to model a private-office threshold from observed private-office demand. Every statement above about dedicated space rests on the weekly-demand proxy, not on observed office take-up, and the workspace-type table should be read as a description of coworking and meeting-room behaviour.

Cluster boundaries were chosen after inspecting the distributions rather than before. Booking days per active market-month run 1 at the 10th through 30th percentiles, 3 at the median, and 13 at the 90th, with a large spike at exactly one day, which is why the one-off and trial groups are split on days rather than months. Survival is measured as any booking in or after a given month, so a market that returns after a gap counts as alive; a stricter contiguous definition would produce lower numbers.

The predictor tables are descriptive, not causal. A market with a busy first quarter differs from a quiet one in ways this data cannot see, including whether a manager was pushing adoption or an office already existed nearby. The rules are useful because the base rates are stable and the negative case is strong, not because early booking days cause later demand. Lead-time figures are reported only for meeting rooms, where a median booking is made 4.6 days ahead and the 90th percentile 20.7 days ahead across 563 bookings; coworking visits are recorded at check-in rather than at reservation, so no meaningful lead time exists for them and none is claimed.

Finally, this describes companies using Croissant, which is a selected population that has already chosen to buy flexible workspace centrally. A market that looks quiet here may be busy in a company office down the road. These figures describe the maturation of flexible workspace demand in a city, which is the relevant input when the question is whether to convert that demand into a fixed commitment. For the employee-level view of the same behaviour, see how often people actually need workspace, and for what happens to supply as these markets accumulate, how distributed companies end up managing 183 operators.