AI is in the building, mostly writing things
Access has broadened quickly: 81% of organisations now authorise employees to use AI, up from 74%. What they do with it is narrower than the headlines suggest. The two leading uses are generating text and content (41%) and writing up meeting minutes (37%).
The third most common answer is the revealing one. Nearly a third (31%) say they simply don't know how AI is being used in their day-to-day work. Access has arrived ahead of understanding.
What is holding it back
The barriers are not really about technology. The top two are lack of expertise (55%) and data quality or availability (51%) — a skills problem and a housekeeping problem. Cost comes third at 36%; staff resistance is well behind at 23%.
Readiness reflects that. Only about a third (34%) call themselves somewhat ready for advanced AI, and the largest group (38%) sits on the fence — neither ready nor unready. What they ask for is unglamorous: ongoing support (77%) and technical training (72%).
- Work that looks polished but has no substance, produced quickly with AI and passed on unchecked.
- It does not save time, it moves it. The person receiving it has to interpret, correct and rewrite.
- The cost is trust as much as hours: recipients report confusion and frustration, and stop relying on the sender.
- The fixes are managerial, not technical — model good use from the top, train people on prompting and review, and treat AI as a second opinion rather than the only one.
The plumbing is better than the intelligence
While AI is still finding its feet, the data infrastructure underneath it has quietly matured. Nearly every organisation now runs a space data management system and feeds it automatically, and outbound feeds to downstream systems have grown sharply.
The stated priorities for that technology are consistent: optimise the portfolio (80%), improve the accuracy of space data (70%) and improve the workplace experience (61%).
That last line is the one to sit with. Half of organisations have no way of telling an employee who will be in the office on a given day — the single piece of information most likely to make the trip worth taking.
Accuracy comes from auditing
Data quality is managed the old-fashioned way: 70% govern it through regular audits, the most common answer since 2023, and 35% audit monthly. Automation is finally arriving — semi-automated audits have nearly doubled to 34%, while fully manual audits fell from 73% to 59%.
But 15% still have no data governance at all, up six percentage points on last year. That is the group most exposed when AI is pointed at their numbers.
- Fix the data before buying the AI. Expertise and data quality are the top two barriers; cost is a distant third.
- Train people, then give them tools. Ongoing support (77%) and technical training (72%) are what organisations say they need.
- Guard against workslop. Require human review, and judge AI output on substance rather than polish.
- Tell people who will be in. Half of organisations still cannot, and it is the cheapest way to lift attendance.
Based on CBRE's 2026 Global Workplace & Occupancy Insights, Part 3, by Susan Wasmund and Lynn Patzner. The study draws on five years of benchmarking and sentiment surveys covering 28 million m² of client portfolios.