The Hidden Tax on AI in Real Estate — And How to Eliminate It

The PropTech Connect webinar brought together a group of real estate technology practitioners for an honest conversation about where AI actually stands today. iREMS Co-Founder Andreas Kozma was among the panellists, and the exchange confirmed something we have been seeing across the market for some time: adoption is accelerating, but institutional trust for high-stakes decisions is still thin — and the gap between organisations that make AI work and those that don't is wider than most are willing to admit. The reason for that gap is almost always the same. It comes down to data architecture.

The Proof Points Are Real — But So Is the Variance

Real estate companies are using AI to cut lease abstraction from days to hours. Portfolio managers are running analysis in real time that used to take a week. These are operational realities at serious organisations, not experiments. But the discussion made clear: implementation quality varies enormously, and success stories often obscure how much invisible cost sits underneath them.

One observation that resonated across the discussion: AI ROI is genuinely hard to measure, not because the speed gains aren't real, but because quality validation is still an unsolved problem for most organisations. Doing something in two hours instead of three days is a compelling headline. Whether the output is better is a different question entirely. The companies being honest about this distinction are the ones building durable advantages.

A Common Pattern: AI as a Bridge Between Systems

A pattern came through clearly in the webinar discussion, and it mirrors what we see across the market: many organisations find themselves turning to AI to manage data fragmentation rather than resolve it. This is understandable — legacy system migration is expensive, disruptive, and rarely straightforward. The pressure to show AI results quickly is real.

The typical starting point looks like this: five, eight, twelve systems — ERP, property management software, valuation tools, spreadsheets, legacy databases — each with its own structure and logic, not designed to talk to each other. AI agents get deployed to act as translators and bridges between these silos. It looks like integration. It feels like progress. And as a transitional measure while a longer-term data strategy takes shape, it can serve a purpose.

The challenge is when it becomes the strategy rather than the bridge to one. AI working across inconsistently structured data is not operating on a reliable foundation — every translation introduces potential error, every upstream system change creates downstream fragility, and the ROI of AI investments gets diluted because a significant portion of the capability budget goes to infrastructure work rather than analytical output.

AI working on fragmented data is like a brilliant analyst whose entire day is spent reformatting spreadsheets before they can start the actual work. The capability is there. The setup is limiting it.

This is not a criticism of the organisations facing it — most inherited these setups long before AI was on the agenda. It is a structural challenge the whole industry is navigating, and the conversation reflected that honestly. What the discussion did confirm is that the deployments producing the clearest results share a common characteristic underneath them: a consistent, governed data foundation.

The Multiplier Effect of Unified Data

The alternative is structurally different, not just technically better. When you build — or migrate to — a unified data model purpose-built for real estate, AI stops being a bridge and starts being an engine.

This is the principle behind iREMS and our Real Estate Unified Data Model^™ (REUDM^™). REUDM™ provides a single, governed, real estate-native data environment where financial, asset, portfolio, and market benchmark data share a common structure and a common language. There are no translations. There are no bridges. The data is already in a form that AI can reason about directly.

The effect on AI performance is not incremental — it is multiplicative. The use cases that produce the strongest results share a consistent characteristic: the AI was given a clear, precise objective and fed structured, validated data. It was constrained by a known data model, not left to infer meaning from heterogeneous inputs. Output quality, iteration speed, and result reliability all improve dramatically when the data foundation is right.

When the data structure is right, the cost-benefit ratio changes. Every AI investment delivers more. Every improvement compounds.

When the Foundation Is Right, the ROI Math Changes

The cost-benefit calculation for AI integration looks very different depending on your starting point. If you are building AI on top of fragmented systems, a significant portion of your AI budget goes to data wrangling, error correction, and bridge maintenance — work that produces no analytical output and must be repeated every time something changes upstream.

When you start from a unified data structure, almost all of your AI spend goes to what AI is actually good at: identifying patterns, generating insights, automating analysis, surfacing anomalies. The ROI curve is steeper and more durable. It compounds because every improvement to the model benefits all use cases built on top of it — not just the one it was built for.

The practitioners seeing the clearest returns are not necessarily running the most sophisticated AI. They are running AI on the best data. Change management, internal champions, and clear objectives all matter — the discussion was emphatic about this — but none of them substitute for structural data quality. They amplify it.

Treat AI Like a New Employee — But Give It a Real Workplace

One framing that came up in the discussion: treat AI like a new hire. A talented new employee dropped into a disorganised office — where records are inconsistent and every team uses different conventions — will underperform regardless of their capability. Give that same person a well-structured, well-documented environment and their output scales.

REUDM^™ is that environment for AI in real estate. The model doesn't change. The workplace does. And the workplace is what determines whether AI delivers on its promise or spends its time reformatting spreadsheets.

AI Adoption Is a Transformation Project — Treat It Like One

One aspect that tends to get less attention in the AI conversation is the methodology — not just the technology stack, but how the transformation is actually run. Vendor selections, model comparisons, and API integrations dominate the discussion. But the organisations consistently getting the strongest results seem to have one thing in common beyond tooling: they treated the change as an organisational project, not a software rollout.

At iREMS, every implementation has always started the same way — not with software configuration, but with a structured process of conceptual discussion, business process analysis, and change management. Before a single data field is mapped or a workflow is built into the platform, we work with the client to understand how information actually moves across their portfolio: how decisions are made, where the friction sits, what the real source of truth should be, and how people at every level relate to the data they use every day. That process is what builds the roots of a unified financial data structure that will actually hold.

The methodology that builds a unified data foundation is the same methodology that makes AI adoption stick. The work is not different. The urgency is.

AI adoption requires exactly this. It is not a tool you bolt on to an existing process. It is a transformation of how an organisation generates, validates, and acts on information. That transformation only holds when the underlying workflows are correct, the people understand why a unified data model matters, and the processes have been redesigned around a single source of truth rather than inherited from legacy habits.

This is the parallel that does not get enough attention: the process iREMS follows in every implementation — conceptual alignment, process analysis, data governance, workflow redesign — turns out to be the prerequisite for AI readiness. Not by design originally, but because both transformations require exactly the same foundations. The organisations that have gone through this process — that have standardised their data, aligned their processes to industry best-practice workflows, and built the organisational discipline to trust a governed data environment — are precisely the ones best positioned to capture AI's compounding returns. They have already done the hard work, even if they did not frame it that way at the time.

For organisations earlier in that journey, the path to AI readiness is also the opportunity to build the foundation properly — and doing both together is more efficient than sequencing them. What we have found is that having a partner who understands both the data model and the organisational change practice makes a meaningful difference. The methodology we have developed and refined across international portfolio implementations maps closely onto what AI adoption requires. That is not a coincidence — it is the same underlying transformation, applied to a new context.

The Compounding Advantage

The most important takeaway from the PropTech Connect webinar conversation— and from everything we see in the market — is this: the gap between organisations that invest in unified data infrastructure now and those that patch their way forward with AI bridges is not linear. It compounds.

Consider the wider context. Recent AI cost benchmarks show that the cost of a unit of AI intelligence has fallen 150x in three years while capability has nearly tripled. The conclusion across SaaS and technology analysts is consistent: this creates the most asymmetric opportunity since the shift to cloud, and the winners will be the companies with distribution, brand, and fundamentals — not wrappers riding the hype.

For real estate, the implication is direct. As AI gets cheaper and more powerful, access to the technology stops being the differentiator — it is already commoditising. The organisations that capture this opportunity will not necessarily be the ones that move fastest to adopt AI. They will be the ones whose data foundation means every new AI capability they adopt immediately has something reliable to work with. A clean, unified, governed data environment is the fundamental that determines whether falling AI costs translate into compounding returns or into cheaper versions of the same fragmentation challenge.

When AI intelligence costs fall 150x, the moat is not the AI. It is the data foundation that makes the AI useful.

Every month of clean, structured data is a month of better AI signal. Every validated workflow built on a unified model is a baseline that improves with each iteration. Every year spent maintaining bridge architectures is a year of technical debt compounding on a fragmented foundation — regardless of how capable or affordable the AI on top of it becomes.

The teenager phase of proptech AI is over. The companies that win the next decade will not necessarily be the ones with the most AI. They will be the ones whose data architecture lets AI do what it was built to do. Conversations like the one at the recent PropTech Connect webinar — and the broader market dynamics — make clear the urgency is real.

ire.ms