The assumption worth testing first is that artificial intelligence in a large organization mostly means fewer people doing the same work. Heimstaden, a European residential real estate operator with roughly 156,000 homes across nine countries, offers a case that runs against that expectation — because the work itself changed shape in ways the old org chart could not have predicted. [1] In June 2025 the company’s leadership chose to accelerate its AI bet and settled on OpenAI after evaluating several platforms, with a stated goal of both empowering employees and rethinking core business processes. [1] What followed is a set of concrete gains, documented in specific numbers, that show where the technology lifted output rather than merely substituting for it. The evidence is narrow and operational, and that is precisely what makes it worth reading closely.
The vacancy problem, and what it used to cost
When an apartment empties, the traditional path back to market runs through several distinct stages: assessing rental value, arranging photography, producing advertisements, and responding to demand as it arrives. Each stage consumes time, coordination and specialist input. Underneath all of it sits an information problem — knowing what you own, what competitors offer, and what a prospective tenant actually wants. Heimstaden now applies AI across each of those steps. Pricing recommendations are generated from tools built in ChatGPT Sites. Image styling and apartment descriptions are produced with AI. Listings are reviewed against the company’s quality standards, and market feedback is synthesized to help letting teams make corrections, which in turn makes pricing more dynamic over time. Employees still review and approve every decision; what has shifted is the volume of preparatory work that no longer lands on a desk.
The clearest single figure comes from the Netherlands. Producing a virtually staged property image there once cost roughly €50. [1] With AI, the same image costs less than €0.10 — a reduction of more than 99%. Heimstaden estimates an annual savings potential of €130,000 in that market alone. The gain is not only cheaper advertising. Applying AI across the letting process gives the company a more responsive way to bring homes to market, shortening the distance between a vacancy and a signed lease.
Three pillars, one process
The letting workflow is not an isolated experiment. It is the practical expression of a strategy built on three pillars: empowering employees through training and access to the right tools; running a central transformation office plus a network of AI champions to drive larger initiatives; and strengthening foundations by making property management systems and data models ready for AI. Letting exercises all three at once — colleagues create staged images and listings, the central team coordinates improvements across pricing and advertising, and property data is made accessible to the tools that support the process. As Chief Digital Officer Nikolaj Brammer described it, the company started by breaking down its most important business processes and asking where AI could create value, rather than automating a single task in isolation.
A sustainability analyst builds her own pipeline
Perhaps the most instructive gain happened far from any software engineering team. Heimstaden needed to assess environmental risks across its portfolio, and the necessary information existed but lay scattered across public sources, formats and datasets. Turning that into a usable assessment would normally have demanded substantial manual labor and technical resources. Julia Erlandsen, an environmental sustainability analyst without a software development background, used Codex to build a different route: supporting data discovery, cleaning and standardization, quality checking, and the development of repeatable processing pipelines, while she retained control over methodology and validation. The result was 20,112 assets assessed and more than six million rows of useful data generated. What would have required an estimated five people working for a year was delivered by two people in two months, and the company avoided engaging external consultants it would otherwise have needed. The capability is now repeatable, supporting climate-risk assessment, reporting requirements and investor decision-making.
Adoption as the real bottleneck
Rollout data suggests the constraint on value was never the technology. After ChatGPT Enterprise launched in September 2025, half of employees were using it within two to three months. A central AI transformation office maintains a library of use cases so colleagues can learn from examples elsewhere in the business, and employees work within rules for data security, retaining responsibility for reviewing AI-generated content and approving decisions. Adoption reached real estate caretakers, including some initially skeptical. In Norway, one caretaker used ChatGPT to assess a neglected apartment and draft a prioritized refurbishment plan, converting observations into a structured document that could be reviewed and refined. The pattern across these examples is consistent: the tool does not decide, it organizes the material on which a person decides.
What the numbers actually show
The question that remains is not whether the gains are real — the cost figures and the compressed delivery timelines support that. It is whether the same logic scales beyond letting and portfolio analysis. Heimstaden is already preparing the next stage: a Procurement Intelligence Platform that combines workflow automation with OpenAI models accessed through APIs across Outlook, SharePoint and Teams, and a co-developed property management system, Amplio Estate, built to let employees retrieve property information and act in core systems with human oversight. Each of these keeps a person at the approval step. That is the pattern worth watching — not whether the buildings run themselves, but whether the people running them keep the final word.