Structured data is the precondition, not the product
Most of what is called a data problem in real estate is a structure problem. Leases as scanned PDFs, cost schedules in inconsistent spreadsheets, technical reports without a common naming logic: no system extracts reliable insight from material that was never organised.
The first gain from technology is therefore unglamorous. Consistent naming, one version per document, a defined folder logic and a complete index produce more measurable speed in a transaction than any model applied to the same material afterwards.
Where it genuinely helps: mapping and screening
Identifying which investors are plausibly active in a segment, which owners hold comparable assets, which parties have transacted in a region — this is pattern work across large, messy, public and semi-public sources, and it is where tooling earns its place.
The output is a longer, better-informed starting list. The selection from that list still requires knowing which of those parties is genuinely able to act today, which is rarely in any dataset.
Document review: faster reading, not final answers
Language models are effective at extracting clauses, flagging inconsistencies between documents, summarising long leases and drawing attention to unusual terms. In due diligence that shortens the first pass considerably.
What they do not do is take responsibility for the conclusion. A model that overlooks a break option or misreads an indexation clause does not carry the consequence; the buyer does. Extraction can be assisted, assessment cannot be delegated.
The limits are structural, not temporary
Three limits will not be solved by better models. First, confidential transactions run on information that is deliberately not public and often not written down. Second, the decisive variables — motivation, timing, internal politics of a counterparty — are known through relationships, not records. Third, professional processes carry liability, and liability requires a person who assessed and decided.
This is not scepticism about the technology. It is a statement about which part of the work is being automated: the reading, not the judging.
Confidentiality sets a hard boundary
Every tool that processes transaction documents raises the same questions: where does the material go, who can access it, is it retained, is it used for training, and does its use comply with what was agreed with the owner and the counterparties.
In a mandate governed by non-disclosure agreements these are not IT questions. Uploading a data room into an uncontrolled service is a breach regardless of how useful the output is — which is why the rules described under off-market advisory apply to tooling as strictly as to people.
What this changes for the next few years
The plausible outcome is not automated transactions. It is a shift in where time is spent: less on assembling and reading material, more on structuring, on selecting counterparties and on negotiation.
For owners and developers, the practical implication is unchanged and slightly sharper: organised documentation is now also the condition for benefiting from the tools the other side is already using.
This article describes general market practice. It is not legal, tax or investment advice and does not replace an assessment of the individual case.
