Worked example
What a Minimum Viable Foundry looks like in a REIT.
Asset management as the first wedge: the entities, the sources, the truth profile, the queries, the agent ladder.
A Minimum Viable Foundry is the smallest governed platform wedge that makes one operating domain queryable and agent-ready. Not a roadmap. Not a phased program. The first wedge that works.
Whole-company platforms fail. They fail because every workflow is different, every source has its own truth profile, and the people who have to approve the work cannot approve a platform that promises to eat the entire enterprise. Wedges work because they are small enough to ship and small enough to refuse.
This note walks through what a first wedge looks like for a typical vertically integrated REIT, using asset management performance as the worked example. Other wedges — leasing pipeline, property operations, capital projects, finance close, investor reporting, ESG, vendor intelligence — follow the same shape.
Pick the domain
Asset management performance is a strong first wedge for most REITs. The questions are specific, the entities are well-known to the operating team, the metrics are reviewed monthly, and the failure mode of an answer is visible (a wrong number in front of asset leadership is a wrong number that gets noticed).
The wedge boundary is one cycle of asset performance review across the portfolio. It is not the entire asset management function and not the underlying budgeting process. The wedge ends where the cycle ends.
Canonical entities to model first
For asset management performance, the entity model only needs the objects that the cycle actually touches. Defer everything else.
- property
- portfolio
- tenant
- lease
- rent_roll
- occupancy_record
- budget
- actual
- forecast
- variance
- metric
- definition
- source_system
- source_object
- truth_profile
- document
- permission_policy
- query
- answer
- agent_capability
- trace
Twenty-one entity types. Not the seventy-plus a full REIT ontology contains. We will add debt instruments, capex projects, vendor contracts, and ESG obligations the moment a workflow needs them. Not before.
Sources and their truth profiles
The first wedge connects to one or two source systems and treats the rest as metadata for now. For asset management at most REITs, that means the property management system as the operating record and the finance system as the financial record, with a small set of governed documents.
Property management system (rent roll, occupancy)
Authoritative
Finance system (GL actuals, approved budget snapshot)
Authoritative
Asset plan documents (current cycle)
De facto trusted
Property-level operating review templates
De facto trusted
Manager commentary documents
De facto trusted
Forecast spreadsheet (regional)
Fragile
Same-store basis adjustments
Fragile
Disputed occupancy adjustments at quarter-end
Disputed
The forecast spreadsheet and same-store adjustments are flagged fragile because they depend on regional judgment and manual edits. The disputed occupancy adjustments are real: most REITs have property-level adjustments that finance and operations argue about every quarter. The platform does not pretend that disagreement does not exist; it tags it.
An agent that answers asset performance questions on this wedge may cite the rent roll and the GL. It may not cite the forecast spreadsheet without flagging it as fragile. It may not act on the disputed adjustments without escalating.
The queries the wedge can answer
Before any agent is built, the wedge supports specific, evidence- grounded questions. Each answer carries source references, freshness, the definition used, confidence, and an escalation path.
- “Which properties are below budget this quarter, and why?”
- “Which assets have conflicting occupancy figures across reports?”
- “Which leasing assumptions changed since the prior operating review?”
- “Which capital projects are over budget and missing owner commentary?”
- “Which lease expirations are material in the next two quarters?”
- “Which operating metrics are disputed, stale, or manually adjusted?”
- “What documentation is required for the next investor reporting package, and what is missing?”
These are not free-form prompts to a chatbot. They are typed queries against the governed semantic layer, returning structured answers an asset manager can take into a review.
The first agent capability
Once the queries above are reliable, one agent capability rides on top: a variance-explanation drafting agent. It pulls the property, the budget, the actual, the variance, the relevant metric definitions, the manager commentary if it exists, and drafts a variance explanation for human review.
The agent operates at rung 5 of the safety ladder — drafts with human approval. It cites every source. It refuses on disputed adjustments and routes to the asset manager. It escalates when the variance crosses a materiality threshold and the manager commentary is missing. It does not send anything externally and does not modify any system of record.
What is deliberately out of scope
The first wedge does not cover debt, capital projects beyond budget variance, ESG, investor relations, leasing pipeline, vendor intelligence, finance close, or compliance. It does not cover write-back to any operating system. It does not cover external communication with tenants, lenders, or investors. It does not attempt to reconcile the disputed adjustments; it surfaces them.
Every one of those is a candidate for the second or third wedge. The decision to add a wedge is a separate decision under change control, on the strength of evidence from the first one.
Why this works
Generalize from a wedge that works. Not from a roadmap that does not. The team building wedge two inherits the source inventory, canonical model, semantic layer, permission model, query layer, observability, and operating record from wedge one. Each wedge compounds. The platform grows by accretion of governed reality, not by ambitious blueprint.
That is the entire bet. The first wedge has to ship, has to pass security review, has to answer specific questions with citations, and has to do all of that without bypassing your existing systems. Once it does, the next wedge is no longer a leap.
More perspectives
Thesis
Most enterprises don't have an AI-agent problem. They have an operating-intelligence problem.
AI maturity surveys are flat. Pilots run hot, production stalls. The model is rarely the issue. Fragile truth is.
Read noteMethod
Five readiness paths: how to recommend not building.
Build is one of five outcomes from a serious AI diagnostic. Three of the others are forms of saying no, on purpose.
Read note