Where Humans Stay in Charge of AI in Real Estate
Five newsroom essays on the judgment calls, accountability and oversight that stay with people as AI takes on more work.
Frederik Raspé, CEO & Co-Founder ·
The Irreplaceable Role of Human Judgment in Real Estate Investing
There is a tempting story in AI strategy that says the answer to bad decisions is more data, better models, and fewer humans in the loop. HBR recently criticized that logic as a form of 'dataism', the mistaken belief that enough information and algorithmic processing will reveal truth automatically. That critique is especially relevant in real estate investing, where ambiguity, incentives, timing, and qualitative judgment remain central.
Real estate decisions are rarely just analytical. They involve interpretation. Two assets with similar metrics can imply fundamentally different opportunities because of local knowledge, sponsor quality, tenant dynamics, political context, or strategic fit. A 5,000 sqm office in Munich and a 5,000 sqm office in Düsseldorf may look comparable on a spreadsheet, but the experienced investor knows they are not.
HBR's critique of dataism applies directly here: the belief that more data automatically produces better truth ignores the role of context, imagination, and moral judgment. In CRE, those capabilities are not peripheral. They are often what separates good decisions from merely data-consistent ones.
AI can support these judgments. It can improve visibility, consistency, and option generation. But it does not eliminate the need for the human capabilities that make real estate investing a judgment discipline, not just a data processing exercise.
Data helps when the relevant variables are known, measurable, and stable. In real estate, many of the most important variables are none of those things. Market sentiment shifts. Regulatory environments change. Tenant creditworthiness depends on factors outside the lease. Sponsor reliability is a relationship judgment, not a data point.
The danger is not that firms use too much data. The danger is that they confuse data coverage with decision quality. A model that ingests every available metric can still produce a confidently wrong answer, because the decisive factor was something the model could not measure.
This is not an argument against AI. It is an argument for using AI to strengthen the moments where human judgment matters, rather than assuming that human judgment is the problem to be solved.
What Should Humans Still Own in an AI Native Investment Process?
As AI capabilities improve, firms need a clearer way to think about ownership. The useful distinction is not 'human versus machine.' It is which layers of work should remain human owned even when machine support becomes extensive. HBR and MIT SMR both point toward the same answer: the more a task involves values, exceptions, accountability, and nonstandard trade offs, the more important human ownership becomes.
The question 'Will AI replace this task?' is being asked in every CRE firm. It is the wrong question. The better question is: 'When AI can support this task extensively, who should still own the outcome?'
HBR's recent work on human decision making and MIT SMR's writing on decision rights converge on a useful framework: machines can increasingly help with collection, normalization, summarization, scenario generation, and recommendation. But humans should still own the goals, the thresholds, the value judgments, the handling of unusual situations, and the final responsibility for outcomes.
Making this distinction explicit is what separates controlled AI adoption from drift.
These three levels are often conflated, which creates confusion about accountability. Assistance means AI helps a human do their work faster, the human remains fully responsible. Recommendation means AI proposes a course of action, the human decides whether to follow it. Ownership would mean AI makes the decision and bears the consequences.
In CRE investment processes, most current AI operates at the assistance level. The strategic question is how to expand into recommendation, where AI surfaces options and trade offs, without accidentally sliding into effective ownership, where humans rubber-stamp machine outputs without meaningful review.
The distinction is not academic. It determines liability, governance, and institutional learning.
How to Design Human in the Loop Workflows for Real Estate AI
'Human in the loop' is one of the most overused and least operationalized phrases in enterprise AI. MIT Sloan Management Review's recent work on agentic AI and decision rights makes the stronger point: once intelligent systems begin shaping choices and actions, organizations must redesign responsibilities, escalation paths, and accountability structures, not just add a review checkbox.
In many organizations, 'human in the loop' simply means that a person glances at the output before it goes live. That is not governance. It is a ritual that provides the appearance of oversight without its substance.
MIT SMR's work on accountability for agentic AI argues that once systems begin shaping choices and taking action, the design of human oversight must become much more deliberate. Where does review happen? What triggers escalation? Who owns the outcome when the machine was wrong?
In CRE, these questions matter enormously, because the stakes of real estate decisions are high, the context is often ambiguous, and the consequences of a bad output can be long-lasting and expensive.
The problem with unstructured oversight is that it creates the illusion of control. When a human reviews an agent's output without clear criteria, time pressure and cognitive load almost guarantee that the review becomes a formality.
This is especially dangerous in high volume workflows. If an agent produces twenty summaries and a human approves them in fifteen minutes, the effective review per document is measured in seconds. That is not oversight. It is rubber-stamping.
HBR's writing on managing work around AI agents reinforces this: the challenge is not adding humans to the loop, it is designing the loop so that human intervention actually changes outcomes.
How Leaders Should Prepare CRE Organisations for Agentic AI
Preparing for agentic AI is not a one-department exercise. HBR's recent work on talent needs and McKinsey's writing on agentic organizations both make that clear: firms will need to rethink workforce design, governance, technology foundations, and management practices together, not sequentially.
Many CRE leaders treat AI adoption as a technology decision: select the right vendor, configure the tool, train the users. HBR and McKinsey both argue that this framing misses the larger challenge. Agentic AI does not just add a capability to the existing organization. It changes what the organization needs to be good at.
That means preparing for agentic AI requires parallel work streams: identifying where agents can add value, redesigning roles and workflows, building data foundations, and establishing governance frameworks. Doing these sequentially, technology first, organization later, is the pattern that produces stalled pilots.
The firms that move well will not chase every use case at once. They will choose a few workflow areas where improved speed, structure, and traceability matter immediately, then build from there.
The most common AI failure mode is not model quality. It is organizational absorption. The technology works, but the firm's workflows, role definitions, and management expectations were not updated to accommodate it.
HBR's writing on workplace redesign makes this explicit: agentic AI requires new role definitions, new spans of control, new management practices, and new performance expectations. These are organizational design problems, not IT problems.
For CRE firms, this means that the CEO, COO, and heads of investment must be involved in AI strategy, not just the CIO or innovation lead. The decisions that matter most are about how work gets done, not which technology gets purchased.
Better Decisions in the Age of AI Start With Better Choices
Most executive AI discussions focus on faster decisions. That is understandable, but incomplete. MIT Sloan Management Review recently made the case clearly: organizations often do not suffer because they cannot decide, they suffer because they are deciding between weak, incomplete, or badly framed options. AI's real contribution may be less about replacing decision makers and more about improving the quality of available choices.
Speed is the dominant promise of enterprise AI. Faster extraction. Faster analysis. Faster reporting. Faster everything. And speed matters, but it is not the same as quality.
MIT Sloan Management Review's recent research on intelligent choice architectures argues that the raw material of good decisions is not data or velocity. It is the quality of the choices presented. When decision makers are offered better-framed options, clearer trade offs, and more relevant comparisons, their outcomes improve, regardless of whether those decisions happen faster.
In CRE, this reframe is especially important. Bad investment decisions rarely stem from a total absence of information. They more often come from narrow framing, poor comparison sets, weak visibility into alternatives, or organizational bias toward familiar templates.
The speed narrative is seductive because it is easy to measure. A report that took three days now takes thirty minutes. A rent roll extraction that required four hours now takes four minutes. These are real gains. But they leave the decision architecture untouched.
Consider an investment committee meeting. The speed at which materials are prepared matters. But what determines the quality of the decision is whether the committee is seeing the right comparisons, the right risk factors, the right alternative scenarios, and the right historical precedents. Producing the wrong analysis faster does not create better outcomes.
HBR's work on the limits of pure data driven substitution for judgment reinforces this: more information and faster processing do not automatically produce better truth. The structure of the choices matters at least as much.