At the recent Altus Innovation Summit in Paris, more than half of the programme was dedicated to answering one question: what does artificial intelligence actually mean for commercial real estate once the novelty wears off? A keynote, a panel and a live product demonstration each took a different angle on it, providing experiential recommendations and insights drawn from what's already working, and what isn't, inside real CRE teams.

The keynote: a deliberately contrarian starting point

Vincent Pavanello, co-founder of Mister AI, a French firm specialising in generative-AI consulting and training, opened with a memorable example of how fast AI adoption has moved: he taught his 95-year-old grandmother to use ChatGPT in three minutes flat. He reminded the room that ChatGPT reached roughly 1.3 billion users in three years, without the typical big budget marketing spend behind it, while Anthropic, by his account, passed a billion dollars in monthly revenue in December 2025 alone, with an annualised run rate of twelve billion. No software company, he argued, has grown at that pace before.

But as general-purpose LLMs become more popular and more capable, handling more of the CRE pipeline's everyday tasks, what role remains for specialised tools? "The real challenge is understanding what, in the tools you use, a general-purpose model simply can't replicate," he said. General-purpose models and specialised software aren't competing; in his view, they're complementary. What a tool like Altus's offers, which no generalist model can, is proprietary data and calculations built specifically for the CRE domain.

His guidance to buyers was practical: pick a strong, secure general-purpose tool, train teams properly on it, then choose specialised tools with intent rather than by default. He didn't sugarcoat the confidentiality question either. Shadow AI use (the unauthorized use of unvetted AI tools by employees), he said, already dwarfs official policy, citing one bank where employees were using free ChatGPT roughly a hundred times more than the firm's own mandatory tool. "A lot is happening in AI, maybe what I'm saying will soon be outdated," he admitted. "Everyone is questioning what their software stack will look like tomorrow."

The panel: how far does trust actually extend

The panel, moderated by Daniela Dragoi, Manager of Valuation Advisory at Altus, brought together Vincent Pavanello, Nicolas Le Goff, who leads Altus's valuation advisory business across EMEA, and Jean-Philippe Carmarans, Chair EMEA Valuation & Advisory, and Head of Valuation in France at Cushman & Wakefield .

Dragoi opened with the observation that CRE has moved well past debating whether to adopt AI: it's already embedded in workflows. The question now is how to use it well, and specifically how to make its output defensible in front of a client or an investment committee.

Nicolas zoomed in on that question. It isn't enough for an AI tool to be right, he argued; it has to be right the same way twice. Ask an LLM the same question today and tomorrow, and there's no guarantee the answers will match, which is a real problem for CRE. That reproducibility gap doesn't exist with ARGUS Assist, which doesn’t run CRE calculations through its own LLM; instead, it relies on the ARGUS calculation engine to generate consistent and defensible outputs every time.

Jean-Philippe pointed out that real estate is awash in secondhand information, like media coverage of transactions and prices, but its accuracy is "very, very relative." His firm trains its models instead on proprietary appraisal data, where every asset is broken down by area, quality, and type, with a precision that cannot be rivaled by publicly available data. The payoff has been concrete: AI agents that check completed reports now do in two seconds what used to take two hours, freeing appraisers to spend that time on judgment rather than proofreading.

It’s a conviction Altus shares: real estate’s signal-to-noise problem isn’t solved by having more data, it’s solved by having the right kind. ARGUS Intelligence is built on years of investment-grade DCF valuations, each modeled at the asset level with the same calculation engine that lenders, auditors, and counterparties have relied on for over thirty years. That’s very different from data aggregated through scraped lists or secondhand transaction coverage. When AI is built on quality data, the efficiency gains hold up under scrutiny.

Nicolas added a related point: the productivity gains become undeniable when ‘the marginal cost of additional data becomes ridiculously low.’ An analyst asked to extract twice as many data points takes roughly twice as long; an AI tool costs perhaps ten extra seconds. That near-zero marginal cost, he argued, makes entire categories of analysis worth attempting that were never worth the time before.

ARGUS Assist: the showcase

Matt LaHood closed the session with a live demonstration of ARGUS Assist, Altus's new agentic AI layer over its valuation platform. Rather than opening a model and re-entering data by hand, he typed an address and let the system pull up the asset's prior valuation and roll it forward to the current quarter, carrying over every assumption automatically.

From there, the demonstration turned to a leasing scenario: a vacant industrial property with a speculative lease not expected to sign until the following January. Working entirely through prompts, Matt increased the tenant-improvement allowance and added a month of free rent to see whether better incentives could pull the signing date forward. The model recalculated instantly, showing a roughly $1.8 million increase in value and a proportional lift in cash flow. Once the terms looked workable, a single instruction turned the scenario into a permanent update, complete with a written summary of exactly what had changed.

The demonstration showed speed without sacrificing rigour. Each change ran through the ARGUS calculation engine, with a written summary of exactly what had been altered, but a tenancy schedule update that once took an analyst a day to complete, review, and return to a client now runs in twenty to thirty seconds. That, in summary, is the answer the Summit was convened to find: not AI as a wholesale replacement for professional judgment, but as a precision instrument; most powerful when its limits are understood, its data is trusted, and the people directing it know the difference.

Frequently asked questions

What is the difference between general-purpose AI and specialised AI tools in commercial real estate?

General-purpose LLMs handle a broad range of tasks but lack access to proprietary data and domain-specific calculations. Specialised CRE tools like ARGUS Assist combine AI with trusted, asset-level data and consistent calculation engines, making their outputs defensible in front of clients and investment committees.

Why is reproducibility important for AI in CRE valuation?

An LLM asked the same question on different days may return different answers. For CRE professionals, whose outputs must be consistent and auditable, reproducibility is a baseline requirement, not a nice-to-have.

How is AI changing the productivity of CRE appraisers and analysts?

Tasks that once took hours or days, including report checking and tenancy schedule updates, now complete in seconds. AI also reduces the marginal cost of additional data analysis to near zero, making previously cost-prohibitive analysis routine.

What role does data quality play in AI-driven real estate valuation?

The accuracy of AI outputs is only as good as the data underlying them. Firms relying on publicly available or secondhand information face significant limitations, while those training models on proprietary, structured appraisal data achieve materially more precise and reliable results.

Want to be notified of our new and relevant CRE content, articles and events?