AI for Co-working and Flexible Workspace

Churn prediction from access data. A member whose attendance has been declining for several weeks is frequently already deciding to leave, and in most operations nobody notices until notice is served — at which point the conversation is a retention negotiation rather than a relationship one. Modelling attendance decline against tenure, contract type and support history gives account managers weeks of lead time, uses data you already collect for building security, and needs no new instrumentation. Nothing else in this vertical returns as reliably.

From your own data, in most places, yes — and check the jurisdiction first. The algorithmic pricing ordinances now in force across a growing list of US cities target software recommending prices using non-public competitor data, which puts pricing from your own occupancy, booking and contract history in a different position. Separately, where a price is set for an individual using their personal data, disclosure obligations may attach — New York's requirement has been in force since 10 November 2025. Segment-level pricing avoids most of that. The commercial constraint is usually tighter anyway: members compare notes, and differentiation you cannot explain damages the community you sell.

Measure peak concurrency rather than average occupancy, and measure it by cohort. Your access control system already holds the record, and joining it to contracts tells you what ratio the building actually supports rather than what a rule of thumb suggests. The analysis usually shows that utilisation varies far more by member segment than operators expect, which means a single overselling ratio across the membership is leaving either capacity or member experience on the table.

In our experience, no — not before the churn model. It is a recurring product idea in this sector and it consistently under-delivers against the effort, because members join for space, price, location and convenience, and community forms through proximity and events more reliably than through suggested introductions. It also uses the same member data the churn model needs, so the sequencing is straightforward: build the thing that affects revenue directly, and revisit matching if the data foundation makes it cheap later.

Look at bookings against actual attendance. In most operations rooms appear more constrained than they are because a meaningful share of bookings are never used, which produces member complaints about availability alongside empty rooms. The data to see this sits in the booking system and the access logs, the fix is usually policy rather than modelling — release windows, no-show consequences, shorter default durations — and it improves the member complaint that operators most often try to solve by building more rooms.