Who owns the memory.

80% of companies use AI, 6% see money come out of it. That figure comes from McKinsey research and was highlighted this summer by Aditya Sanghvi, senior partner at QuantumBlack and head of the firm's global real estate practice. In a long conversation, Sanghvi explained why AI is still barely changing results anywhere in real estate and facilities, and what needs to change.

His diagnosis is sharp, and his message for real estate and facilities is sharper still. We see three things in it that really matter for Dutch organisations with a property portfolio or a large facilities operation.

The work is in everything around it

Sanghvi says that in real estate and facilities, the real work isn't the problem. Everything around it is. The time goes into tickets, emails, passing things along and chasing what gets left behind. Plumbers, technicians and property managers aren't the bottleneck. What does hold things up is the dead space between intake, dispatch, execution and feedback.

That's a different way of looking at it. AI doesn't replace skilled professionals. AI clears away everything around their work, and that's where a large share of the cost and lead time sits.

Agents that work together count, standalone helpers don't

According to Sanghvi, the reason the 94% see nothing is simple. Companies use AI to make one employee faster, not to make the whole organisation faster. Someone who writes emails faster with ChatGPT will never deliver significant savings. A chain of agents that takes in a fault report, dispatches it, follows it up and feeds the pattern back into preventive maintenance will.

Those agents work together through the Model Context Protocol, an open standard that almost every serious AI system adopted this year. We build on it. Our recent secure AI assistant for the real estate and facilities department of a Dutch public body works exactly like this, with its own data and its own models in an environment the organisation manages itself.

Who owns the memory

The most important question Sanghvi raises is this. As AI in real estate and facilities increasingly supports or makes decisions, who owns the memory of those decisions? The tool provider, the platform vendor or the organisation itself? That memory (which faults were resolved and how, which renewals worked, which suppliers perform, which interventions really contribute to the bottom line) will soon be the difference between organisations that learn and organisations that merely rent tools.

His advice is crystal clear. Own your data. Don't build your own LLM. Building your own LLM is a billion-dollar undertaking for the likes of OpenAI, Google, Anthropic and Alibaba. You won't beat them and you'll lose focus. What you can do is choose existing LLMs, train them further on your own data and build your own agents and processes around them. That's exactly what makes the difference. Control over your own data and over what your organisation learns is what counts.

What we see in our work

The organisations we work with run into the same three things Sanghvi mentions. The data isn't in order, projects stay too small to change results and leadership often isn't at the table. Our answer isn't a five-year data programme. We tackle the data one topic at a time, such as fault handling, contract management or tendering. That's where the value can be measured immediately and your organisation learns from day one.

Where we land

Real estate and facilities need to get a few things in order now. Most importantly, someone on the board or in the leadership team needs to take this on personally rather than handing it off to IT. Sanghvi is firm about that. If IT leads this programme, you've lost before you start. Next, what counts is whether real estate and facilities own their data and hold on to what the organisation learns, or whether that sits with a vendor. And then there's the choice of which two or three topics to tackle first, where collaborating agents can clear away the work around the work.

The rest follows from there.